CORRECTION: THE "TILE IN VRAM" READING WAS WRONG.  THE EXPORT DOES NOT WRITE IT.

Read VRAM at 0x04017830 after a WHITE export and after a BLACK export, in the
same session:

  fill=ff : 0x04017830 = effefffffdffffff
  fill=00 : 0x04017830 = effefffffdffffff      <-- IDENTICAL

Identical for both fills means those bytes are NOT the exported tile -- they are
ambient VRAM content (the Terminal Manager's own screen), which merely happens to
be mostly 0xFF in that region.  My previous turn reported "the tile is in VRAM"
from a single read of 0xFF bytes; that conclusion does not survive the white/black
comparison and is withdrawn.

Likewise the 18,338 VRAM writes observed "during the export cycle" are the
window-creation / Terminal-Manager repaint, in the same way the earlier 43,882
figure was (see VRAM_WRITE_ATTRIBUTION.txt).  The export's own contribution is
not separable from that traffic and does not carry the tile.

WHAT ACTUALLY STANDS (all reproducible):
  * import verified in the server: element+0x28 = 0xFF white / 0x00 black;
  * export completes without error, under the opt-in VT1200_ACCEL=1 model only;
  * source pixels do reach the accelerator FIFO (0x5600080 writes containing
    0xFFFF / 0x0000) and are staged into main RAM at 0x004FE790;
  * nothing puts them on screen: the 64x64 window is unchanged, and white vs
    black captures differ only by Terminal Manager figures and cursor.

STATE
  * NO client-chosen XIE image has ever been displayed;
  * model opt-in; default configuration unchanged;
  * xshow (core X) remains the only viewer that has shown an image.
